Dataset Card for ogbg-code2

Dataset Summary

The ogbg-code2 dataset contains Abstract Syntax Trees (ASTs) obtained from 450 thousands Python method definitions, from GitHub CodeSearchNet. "Methods are extracted from a total of 13,587 different repositories across the most popular projects on GitHub.", by teams at Stanford, to be a part of the Open Graph Benchmark. See their website or paper for dataset postprocessing.

Supported Tasks and Leaderboards

"The task is to predict the sub-tokens forming the method name, given the Python method body represented by AST and its node features. This task is often referred to as “code summarization”, because the model is trained to find succinct and precise description for a complete logical unit."

The score is the F1 score of sub-token prediction.

External Use

PyGeometric

To load in PyGeometric, do the following:

from datasets import load_dataset

from torch_geometric.data import Data
from torch_geometric.loader import DataLoader

graphs_dataset = load_dataset("graphs-datasets/ogbg-code2)
# For the train set (replace by valid or test as needed)
graphs_list = [Data(graph) for graph in graphs_dataset["train"]]
graphs_pygeometric = DataLoader(graph_list)

Dataset Structure

Data Properties

property	value
scale	medium
#graphs	452,741
average #nodes	125.2
average #edges	124.2
average node degree	2.0
average cluster coefficient	0.0
MaxSCC ratio	1.000
graph diameter	13.5

Data Fields

Each row of a given file is a graph, with:

• edge_index (list: 2 x #edges): pairs of nodes constituting edges
• edge_feat (list: #edges x #edge-features): features of edges
• node_feat (list: #nodes x #node-features): the nodes features, embedded
• node_feat_expanded (list: #nodes x #node-features): the nodes features, as code
• node_is_attributed (list: 1 x #nodes): ?
• node_dfs_order (list: #nodes x #1): the nodes order in the abstract tree, if parsed using a depth first search
• node_depth (list: #nodes x #1): the nodes depth in the abstract tree
• y (list: 1 x #tokens): contains the tokens to predict as method name
• num_nodes (int): number of nodes of the graph
• ptr (list: 2): index of first and last node of the graph
• batch (list: 1 x #nodes): ?

Data Splits

This data comes from the PyGeometric version of the dataset provided by OGB, and follows the provided data splits. This information can be found back using

from ogb.graphproppred import PygGraphPropPredDataset

dataset = PygGraphPropPredDataset(name = 'ogbg-code2')

split_idx = dataset.get_idx_split() 
train = dataset[split_idx['train']] # valid, test

More information (node_feat_expanded) has been added through the typeidx2type and attridx2attr csv files of the repo.

Additional Information

Licensing Information

The dataset has been released under MIT license license.

Citation Information

@inproceedings{hu-etal-2020-open,
  author    = {Weihua Hu and
               Matthias Fey and
               Marinka Zitnik and
               Yuxiao Dong and
               Hongyu Ren and
               Bowen Liu and
               Michele Catasta and
               Jure Leskovec},
  editor    = {Hugo Larochelle and
               Marc Aurelio Ranzato and
               Raia Hadsell and
               Maria{-}Florina Balcan and
               Hsuan{-}Tien Lin},
  title     = {Open Graph Benchmark: Datasets for Machine Learning on Graphs},
  booktitle = {Advances in Neural Information Processing Systems 33: Annual Conference
               on Neural Information Processing Systems 2020, NeurIPS 2020, December
               6-12, 2020, virtual},
  year      = {2020},
  url       = {https://proceedings.neurips.cc/paper/2020/hash/fb60d411a5c5b72b2e7d3527cfc84fd0-Abstract.html},
}

Contributions

Thanks to @clefourrier for adding this dataset.